Live streaming and video

Live streaming, from camera to viewer

Live streaming is a chain: cameras, a switcher, an encoder, a platform and every viewer's player, and each link adds delay or risk. As a hobby I stream and record with Blackmagic ATEM switchers and PTZ (pan, tilt and zoom) and cinema cameras. This page covers video formats, switchers, cameras, cables and distance, lip sync, OBS (Open Broadcaster Software), the protocols that carry streams, multi site shows and the network underneath, and ends with a fault finder.

  • ATEM switchers
  • PTZ and cinema cameras
  • HDMI, SDI and fibre
  • Lip sync
  • OBS Studio
  • RTMP, SRT and NDI
  • Multi site
  • Network
Three cameras feeding a switcher that streams to viewers A PTZ camera, a cinema camera and a second camera send video to a switcher, and a mixer sends sound to it. The switcher sends one programme feed to the streaming platform, which delivers it to phones, laptops and televisions. SDI SRT audio PTZ cameraremote Cinema camstage Camera 3wide shot Switcherprogramme Mixersound Platformingest Viewersphones ViewersTV, laptop
Live view · cameras, switcher, stream and viewers
Stream chain

From camera to viewer, one link at a time

A live stream is a chain. Cameras feed a switcher, the switcher feeds an encoder, the encoder sends one stream to a platform such as YouTube, and the platform delivers copies to every viewer. Sound joins the picture at the switcher. Knowing the chain tells me where delay builds up and where a fault can sit.

The live streaming chain and where delay builds up Five stages left to right: cameras, the switcher, the encoder, the streaming platform and the players. A mixer feeds sound into the switcher. Each stage adds delay: a few frames in the cameras and switcher, under a second in the encoder, seconds in the platform and the most in the player buffer. Camerascapture 1 to 3 frames Switcherprogramme about 1 frame Encodercompress 0.1 to 1 s Platformtranscode 1 to 5 s Playersdecode 2 to 30 s Mixeraudio feed Added delay at each stage (typical) Glass to glass, typical: 3 to 30 seconds with RTMP in and HLS out; under a second with WebRTC.

Swipe sideways to see the whole diagram.

Animation

Orange is camera video, teal is sound from the mixer, purple is the programme stream on its way to the platform and cyan is what viewers receive. Most of the delay a viewer notices sits at the right hand end, in the platform and the player buffer, not in the cameras.

The chain, stage by stage

StageWhat it doesWhat can go wrong
CamerasCapture the picture and send it over HDMI (High Definition Multimedia Interface), SDI (Serial Digital Interface) or the network.A format the switcher cannot take, overheating, a flat battery.
SwitcherChooses which camera is live, adds graphics and keys, and mixes in sound.A wrong input format, a source that will not lock.
EncoderCompresses the programme with H.264, HEVC (High Efficiency Video Coding) or AV1 (AOMedia Video 1) and sends it out.Too little upload bandwidth, an overloaded computer.
PlatformTakes the stream in, makes lower quality copies for slow viewers and sends them out.A wrong stream key, settings the platform rejects.
PlayersBuffer, decode and play the stream on each viewer's device.Buffering when a viewer's connection is slow.

Every stage has to keep up in real time. A recording can be fixed later. A live stream cannot, so I design each link with spare capacity and a fallback.

Video basics

Resolution, frame rate and bitrate

A video stream is a series of frames, compressed so it fits down an internet connection. Three numbers decide most of the result: the resolution, the frame rate and the bitrate. The codec decides how much quality each bit buys. Step through to see what YouTube recommends and what that means for the upload.

Recommended live bitrates by resolution A bar chart of the bitrates YouTube recommends for live streams with H.264, in megabits per second. 720p30 is 8, 1080p30 is 14, 1080p60 is 17, 1440p60 is 34 and 2160p60 is 50. HEVC or AV1 need less. A dashed line shows the upload I plan for 1080p60 with headroom. 10 20 30 40 50 Recommended bitrate, Mbit/s 8 6 720p30 14 10 1080p30 17 12 1080p60 34 24 1440p60 50 35 2160p60 upload 26 Audio: AAC, 128 kbit/s stereo H.264 HEVC or AV1

Swipe sideways to see the whole diagram.

Step 1
  1. YouTube recommends these H.264 bitrates for each resolution and frame rate. H.264, also called AVC (Advanced Video Coding), plays on almost everything, so it is my default.

  2. HEVC and AV1 (AOMedia Video 1) reach the same quality at roughly 30% less bitrate. They need an encoder and a platform that accept them.

  3. I plan the upload at about one and a half times the stream bitrate (typical), so 1080p60 at 17 Mbit/s needs a line that sustains about 26 Mbit/s up.

  4. Sound is small next to video: AAC (Advanced Audio Coding) at 128 kbit/s for stereo is YouTube's recommendation.

The settings I start from

SettingWhat I useWhy
Frame rate25 or 50 fps (frames per second) in the UK; 30 or 60 where the audience and cameras expect itUK mains is 50 Hz, so 25 and 50 avoid flicker from lighting. Every camera, the switcher and the stream must match.
Resolution1080p for most streams; 4K only when the switcher, the upload and the audience can all take itA sharp 1080p stream at a healthy bitrate beats a starved 4K one.
ScanProgressive (the p in 1080p)Interlaced video (1080i) shows combing on phones and computers.
Rate controlCBR (constant bitrate)Platforms expect a steady stream and size their buffers for it.
Keyframe interval2 secondsYouTube recommends 2 seconds and no more than 4, so players can join and switch quality quickly.
CodecH.264 by default; HEVC or AV1 when every part of the chain supports itNewer codecs save bitrate but cost more processing.
AudioAAC, 128 kbit/s stereo on the stream; 48 kHz through the productionYouTube lists 44.1 kHz for stereo. 48 kHz, the usual rate for video kit, also works in practice. I keep one rate from the mixer onwards.

Bitrates and stream settings from YouTube's live encoder guidance, checked in October 2026. Other platforms publish similar figures.

Switchers

The switcher, the heart of the show

A vision switcher chooses which source is on air, adds graphics and mixes in sound. Blackmagic's ATEM switchers do this in a small box, with a streaming encoder and a recorder built into many models. Choose a mode to see how the switcher handles it.

Inside a vision switcher Four sources, three cameras and a slides laptop, feed a switcher with a preview bus, a programme bus, keyers and a multiview output. The programme output goes to the stream and the recorder. A multiview monitor shows four source tiles above a preview tile and a programme tile. Switcher Cam 1wide Cam 2presenter Cam 3PTZ Slideslaptop Preview busnext shot Programme buson air Keyersgraphics on top Multiview Programme outstream and record C1 C2 C3 S PVW PGM Cut: Cam 2 goes from preview to programme at once Mix: Cam 1 dissolves into Cam 2 over about a second Key: slides or a lower third layered over the presenter Multiview: every source, preview and programme on one screen

Swipe sideways to see the whole diagram.

Cut. I set the next shot on the preview bus, check it on the multiview, then cut. The preview shot swaps onto programme in one frame, and the old programme shot moves to preview.

Mix. A mix, or dissolve, blends the two shots over a set time, typically about a second. I keep transitions simple and consistent; the audience should notice the speaker, not the effects.

Key. Keyers layer one source over another: a logo, a lower third with a name, or slides beside the presenter in a box. Upstream keyers sit in the shot; downstream keyers sit on top of everything.

Multiview. One monitor shows every input, the preview and the programme, with a red border on what is live. On an ATEM switcher this is the screen I watch all show long.

Animation
How keyers stack, layer by layer Five layers stack up from the bottom. First the background camera, then an upstream key with slides in a box. Above them sit a downstream key with a logo, a second downstream key with a lower third name strap, and fade to black on top. On the right, the picture the viewer sees builds up layer by layer. BackgroundCam 2, the presenter Upstream keyslides in a box (DVE) Downstream key 1logo bug Downstream key 2lower third with a name Fade to blackstart and end of show slides logo Speaker name, role What the viewer sees

Swipe sideways to see the whole diagram.

Step 1
  1. The programme starts with one source, here the presenter on camera 2.

  2. An upstream keyer places the slides in a box beside the presenter, using a DVE (digital video effect). It belongs to the shot, so it leaves when I cut away.

  3. A downstream keyer puts the logo on top of everything. It stays put whatever I cut to.

  4. A second downstream keyer adds a name strap. I take it on when the speaker starts and off after about eight seconds.

  5. Fade to black takes the whole programme, keys included, to black at the start and end of the show.

The ATEM range, in short

ATEM familyInputsWhat sets it apart
ATEM Mini and Mini Pro4 HDMIThe Pro streams over Ethernet with RTMP (Real Time Messaging Protocol) or SRT (Secure Reliable Transport), and records the programme to a USB (Universal Serial Bus) drive.
ATEM Mini Pro ISO4 HDMIAlso records each input as its own file, called ISO (isolated) recording, and saves a DaVinci Resolve project for editing later.
ATEM Mini Extreme and Extreme ISO8 HDMIMore keyers, a SuperSource for several boxes on screen and two USB outputs. The ISO models record every input, and the newer ISO G2 adds XLR microphone inputs.
ATEM SDI models and Television Studio HD84 to 8 SDISDI inputs for long runs, talkback, and tally and camera control sent to Blackmagic cameras over the programme return.
ATEM ConstellationMany SDILarge broadcast switchers with several M/E (mix effects) rows for big multi camera shows.

Input counts and features from Blackmagic Design's published specifications, October 2026. I use ATEM switchers from both the HDMI and the SDI families.

The same ideas apply to any vision switcher, hardware or software: preview and programme, transitions, keys and a multiview. Software switching in OBS is covered further down.

Cameras

Cameras that cut together cleanly

In my own time I stream and record with PTZ and cinema cameras. Each type has a job: PTZ cameras for fixed angles under remote control, cinema cameras where the picture has to look its best. What matters most is that every camera matches the switcher's format and looks the same when I cut between them.

Camera video, programme return and PTZ control A cinema camera on stage and two PTZ cameras send video to a switcher. The switcher sends its programme back to the cinema camera over SDI, carrying tally, talkback and camera control. A PTZ controller moves the PTZ cameras through a network switch with VISCA over IP. The switcher's programme goes on to the stream and the recorder. programme return: tally, talkback, control VISCA over IP, UDP 52381 programme Cinema cameraSDI, on stage PTZ camera 1video and IP PTZ camera 2video and IP SwitcherSDI inputs Network switchcontrol VLAN Stream and record PTZ controllerjoystick, presets

Swipe sideways to see the whole diagram.

Animation

Orange is camera video into the switcher. Purple, on the dashed line, is the programme return to the camera, which carries tally (red) so the presenter and operator know when that camera is live. Teal is pan, tilt and zoom control from the joystick to each PTZ camera over the network.

Choosing cameras

Camera typeWhere it fitsWhat I watch for
PTZ (pan, tilt and zoom)Fixed positions such as the back of a hall or a meeting room, run by one person from a joystick, with stored presets.Control over VISCA (Video System Control Architecture) on serial or IP (Internet Protocol), or NDI (Network Device Interface); slow, smooth moves on air.
Cinema cameraThe main presenter or performer shot, where picture quality matters most.Heat on long shows, power, and a clean HDMI or SDI output without on screen menus.
CamcorderRoaming shots and long events, with a long zoom and an all day battery.Autofocus behaviour and the output format menu.
Mirrorless or DSLR (digital single lens reflex)Low cost extra angles with good lenses.Clean HDMI output, recording time limits, overheating and fragile micro HDMI ports.
Webcam or phoneRemote guests and quick extra views.Variable frame rate and auto exposure that drifts.

Habits that make cameras cut together

Match the format

Every camera must output the switcher's format, for example 1080p50. A camera set to 1080p25 or 1080i will show black or flicker on many switchers.

Lock exposure and colour

I set white balance, shutter and iris by hand, so shots match when I cut between cameras. Shutter at twice the frame rate, 1/50 for 25 fps, avoids flicker under UK lighting.

Use the return feed

On SDI switchers the programme return gives each camera operator tally and talkback, and lets me match cameras from the switcher with camera control.

Presets save the show

On PTZ cameras I store a preset for every shot I plan: lectern, panel, audience, wide. A preset recall moves the camera in a second, ideally while it is off air.

Cables and distance

HDMI, SDI and fibre, and how far each goes

The cable between a camera and the switcher decides how far apart they can be. HDMI suits a desk, SDI suits a hall, and fibre or the network suit a building or a campus. Choose a link type to see its reach, then try a run of your own.

How far each video cable reaches A distance scale from 1 metre to 10 kilometres, drawn on a logarithmic axis, with a bar for each way of carrying video. Passive HDMI reaches about 10 metres and active optical HDMI about 30. 3G-SDI on RG6 coax reaches about 80 to 100 metres, and 12G-SDI on 12G rated coax about 30 to 90. SDI over fibre reaches kilometres. One copper Ethernet run reaches 100 metres, and network video extends that over any distance. 1 m 10 m 100 m 1 km 10 km HDMI, passiveup to about 10 m at 1080p HDMI, active opticalabout 10 to 30 m 3G-SDI on RG6 coaxabout 80 to 100 m 12G-SDI on 12G coaxabout 30 to 90 m SDI over fibrekilometres, set by the SFP Ethernet, one copper run100 m per run NDI or SRT over the network: any distance Distances are indicative. Cable quality, connectors and the equipment at each end decide the real limit.

Swipe sideways to see the whole diagram.

HDMI is made for a few metres between a box and a screen. Passive cables carry 1080p for roughly 5 to 10 m; 4K needs shorter or active optical cables. The connector has no lock, so I tie it off at both ends.

SDI uses 75 ohm coax with locking BNC (Bayonet Neill Concelman) connectors, built for broadcast runs. Belden's calculated figures are about 78 m for 3G on its 1694A cable and 93 m for 12G on its 4794R. Barco recommends staying under 50 m and 30 m on some cables.

Fibre converters turn SDI into light and back, in both directions on one unit. Single mode fibre with the right SFP (small form factor pluggable) optic carries SDI for kilometres, and fibre cannot pick up hum or interference.

Network video such as NDI or SRT goes as far as the network does. One copper Ethernet run is limited to 100 m, but switches, fibre and the internet extend it, at the cost of compression and some delay.

Animation

Try it: which cable for this run?

Pick the camera output and the distance from the camera to the switcher. The tool suggests the cable or converter I would use. Distances are indicative and depend on the cable and equipment.

40 m
Cable
75 ohm coax, 3G rated or better
Connectors
BNC, locking
Fallback
a spare run, or fibre converters

Run SDI on good 75 ohm coax. At 40 m a 1080p SDI signal is well inside the reach of RG6 type cable. I use cable rated for 3G or better, locking BNC connectors and no adapters along the way.

Reach at a glance

LinkTypical reachNotes
HDMI, passiveAbout 5 to 10 m at 1080p; less at 4KCheap and common. No lock, and the plug is easily pulled out.
HDMI, active opticalAbout 10 to 30 mLight inside, so it is directional: the source end must be at the source.
HDMI over HDBaseTUp to 100 m on Cat6AA transmitter and receiver over one Ethernet cable; covered on the audio visual (AV) over IP page.
3G-SDI on RG6 coaxAbout 80 to 100 mFor 1080p50 and 1080p60. Reclocking distribution amplifiers extend it.
12G-SDI on 12G coaxAbout 30 to 90 mFor 4K. The cable and the BNC connectors must be rated for 12G.
SDI over fibreKilometresConverters at each end with SFP optics; single mode for long runs.
NDI on the network100 m per copper run; any distance across switchesUses the network, so it needs bandwidth and good switching.

Cable figures from Belden's calculated reach charts and Barco's guidance, checked October 2026; real reach depends on installation.

Fibre and copper basics, connectors and testing are on the fibre optics and copper cabling pages.

Audio

Sound for the stream, in sync with the picture

Viewers forgive a soft picture long before they forgive bad sound. Two things matter most: a mix made for the stream, and sound that lines up with the picture. Video takes longer through cameras and converters than sound takes from the mixer, so the switcher has to hold the sound back. Step through a lip sync check.

Lip sync: delaying sound to meet the picture A timeline from 0 to 200 milliseconds. A clap leaves at zero. The picture reaches the switcher at 80 milliseconds and the sound at 5 milliseconds, outside the EBU R37 window. Delaying the sound by two frames moves it to 85 milliseconds, inside the window. 0 ms 40 ms 80 ms 120 ms 160 ms 200 ms Video Audio EBU R37 window picture at 80 ms sound at 5 ms sound delayed 2 frames

Swipe sideways to see the whole diagram.

Step 1
  1. A clap on stage leaves at time zero as sound and picture. It reaches the switcher by two routes.

  2. The camera, its processing and any converter add about two frames, 80 ms at 25 fps (typical).

  3. Sound from the mixer arrives almost at once, 75 ms ahead of the picture. EBU (European Broadcasting Union) R37 asks for sound no more than 40 ms early or 60 ms late.

  4. I delay the sound input on the switcher by two frames. ATEM Mini switchers can delay their analogue inputs by up to 8 frames.

  5. I check with a clap or a sync test clip on the stream itself, because the encoder and platform can shift it again.

Getting sound to the stream

A clean feed from the mixer

I take a dedicated stereo mix for the stream from the mixer, not the room mix. The room needs less of the voices that are already loud in the hall; the stream needs all of them.

Embed or input

Sound joins the picture at the switcher, through its analogue or XLR inputs, or embedded in an HDMI or SDI signal from the mixer's own outputs or a converter.

Levels and loudness

I peak around minus 6 dBFS (decibels relative to full scale) on the switcher meters and never clip. Platforms turn loud streams down; the post production page covers loudness.

Remote guests

A remote guest must hear everyone except themselves, a mix minus, or they hear their own voice back with a delay. I build that on the mixer.

My live sound setup and the mixer side are on the live sound page.

OBS

Software switching with OBS Studio

OBS (Open Broadcaster Software) Studio is free software that does a switcher's job on a computer. It builds scenes from sources such as cameras, capture cards, NDI (Network Device Interface) feeds, images and browser pages, then streams and records the result. Choose a scene to see how its sources stack.

Scenes and sources in OBS Studio OBS Studio in three columns: a list of scenes, the sources inside the chosen scene, and the canvas the viewer sees. The scenes are Starting soon, Main camera, Two shot, Slides and Be right back. Each scene shows a different layout on the canvas, which goes to the stream and the recording. Scenes Sources in the scene Starting soon Main camera Two shot Slides Be right back Holding slideCountdownMusic Camera 1 (NDI)Lower thirdLogoMixer audio Camera 1, left boxCamera 2, right boxBackgroundMixer audio Laptop capturePresenter boxLogoMixer audio Holding slideMusic Canvas, 1920 by 1080 We start soon Speaker name Camera 1Camera 2 slidescam Stream and record

Swipe sideways to see the whole diagram.

Starting soon. A holding slide with a countdown and music, live before the start so the stream is proven and viewers can gather.

Main camera. The presenter from an NDI camera, a lower third for the name, the logo and the mixer's sound. Sources are layered: the top of the list is drawn on top.

Two shot. Two cameras side by side for a conversation, over a branded background.

Slides. The presenter's laptop through a capture card, with the presenter in a small box. Slides need a sharp, steady picture, so I keep the bitrate up.

Be right back. A holding slide for breaks, so the stream never shows an empty stage or a crew conversation.

Animation

The OBS settings I start from

OBS settingWhere it isWhat I set
EncoderSettings, OutputA hardware encoder such as NVENC (NVIDIA's encoder) when the computer has one; x264 software encoding only on a strong processor.
Rate control and bitrateSettings, OutputCBR at the platform's recommended bitrate for the resolution, for example 12 to 17 Mbit/s for 1080p60 H.264.
Keyframe intervalSettings, Output2 seconds.
Base and output resolutionSettings, Video1920 by 1080 for both, so OBS does not scale.
Frame rateSettings, VideoThe same as the cameras: 25 or 50 in the UK.
Audio sample rateSettings, Audio48 kHz, the same as the mixer, to avoid drift.
RecordingSettings, Output, RecordingA local recording at a higher bitrate in a crash safe container such as MKV (Matroska video), remuxed to MP4 afterwards.

NDI sources come into OBS through the DistroAV plugin, formerly called OBS NDI, which needs the NDI runtime installed. vMix and Wirecast are commercial alternatives with built in NDI, instant replay and remote guest calls.

Protocols

RTMP, SRT, NDI and the hop each one suits

Different protocols carry the stream on different hops. RTMP (Real Time Messaging Protocol) carries the finished programme to a platform. SRT carries a camera or a remote site across the internet, NDI moves video around a local network, and HLS (HTTP Live Streaming) takes it from the platform to viewers. Step through how SRT survives a lossy internet link.

How SRT recovers a lost packet A sender at a remote site streams to a receiver across the internet with SRT. One packet is lost on the way. The receiver sends a negative acknowledgement, the sender resends the packet, and it arrives in time because the receiver holds a latency buffer before playing out. Senderremote site Receiverbuffer, then out Internet Receiver asks again for the lost packet (a NAK) The resend arrives inside the latency buffer Latency set too low: no time to resend, a glitch latency buffer

Swipe sideways to see the whole diagram.

Step 1
  1. SRT sends the stream over UDP (User Datagram Protocol), numbering every packet. It can encrypt with AES (Advanced Encryption Standard) at 128, 192 or 256 bits.

  2. The internet drops a packet. Plain UDP would simply lose it, and the picture would break up.

  3. The receiver sees the gap in the numbers and sends a NAK (negative acknowledgement) asking for that packet again. This is ARQ (automatic repeat request).

  4. The sender resends the packet, and it arrives while the receiver is still holding earlier packets in its latency buffer. The output stays clean.

  5. If the latency setting is shorter than a resend takes, the packet arrives too late. I set it to several round trip times; devices commonly default to about 120 ms.

Which protocol for which hop

ProtocolTypical useDelay (typical)Notes
RTMP and RTMPSEncoder to platform, such as YouTube or FacebookA few seconds end to end, set mostly by the platformRuns over TCP (Transmission Control Protocol). RTMPS is RTMP with TLS (Transport Layer Security) encryption, and is what YouTube recommends.
SRTSite to site contribution over the internetAbout 120 ms to a few seconds, set by the latency valueOpen source, from Haivision. Recovers lost packets, encrypts, and works in caller, listener or rendezvous mode.
NDICameras and computers on the same networkAbout a frame or two on a good networkFull bandwidth NDI needs about 125 to 150 Mbit/s for 1080p60; NDI HX3 about 50 to 62 Mbit/s.
WebRTC (Web Real Time Communication)Remote guests and very low delay viewingUnder a secondBuilt into browsers. Harder to scale to large audiences.
HLSPlatform to viewerAbout 6 to 30 seconds; 2 to 5 with low latency HLSVideo sent as short files over HTTP (Hypertext Transfer Protocol), so it scales through content delivery networks.

NDI figures from the NDI documentation and Epiphan; delays are indicative and depend on settings and networks.

Multi site

One show from several locations

Some events happen in more than one place: a second stage, a satellite office or speakers at home. The pictures from each location travel to one main venue, which mixes them into a single programme and sends the show back to every site.

A show built from several locations Two remote sites send their pictures to the main venue over encrypted SRT, and receive the programme back as a return feed. A remote guest joins by a browser call. The main venue mixes everything into one programme and sends it to the platform, which delivers it to viewers. SRT, encrypted programme return WebRTC call Site Bswitcher and cams Site Cone camera Remote guestlaptop Main venueswitcher mixes all Platformone programme Viewers Viewers

Swipe sideways to see the whole diagram.

Animation

Purple is contribution from each remote site into the main venue, cyan dashed is the programme return so each site can see and hear the show, and orange is a remote guest on a browser call. Only one programme leaves for the platform, so viewers see one show.

What makes a multi site show work

One place mixes

Every site sends its pictures to one main venue, which makes the programme. Two switchers both streaming to the platform is two shows, not one.

Return feeds and talkback

Each remote site gets the programme back, so presenters can hear questions, and a talkback line so the director can cue them.

SRT or a bridge

I carry remote sites with SRT, from an encoder or a switcher that supports it, or with NDI Bridge between networks. Blackmagic's ATEM Streaming Bridge turns a stream from an ATEM Mini Pro back into SDI and HDMI at the far end.

Mix minus for every remote

A remote presenter hears the whole show except their own voice. Without that they hear themselves late and lose their place.

Bonded connections

Where a venue has no reliable line, cellular bonding units combine several mobile connections into one, at the cost of more delay.

Rehearse the links

I test every remote link at the same time of day as the show, because shared internet connections behave differently in the evening.

Network

The network under the stream

A modern production is also a network. NDI cameras, the switcher's control, the encoder and the upload all share switches and links, and a stream only stays smooth if those links have room. Choose a camera format to see the load on one uplink, then check a venue's upload.

NDI cameras and the uplink they share Six NDI cameras on an access switch share a 1 gigabit uplink to a production computer. With full bandwidth NDI the uplink is about 90% full; with NDI HX3 it is about 37% full; with multicast and IGMP snooping only the streams in use cross it. Access switch6 cameras, PoE Cam 1 Cam 2 Cam 3 Cam 4 Cam 5 Cam 6 Production PCswitching 1 Gbit/s uplink about 900 Mbit/s, 90% about 370 Mbit/s, 37% only the streams in use cross

Swipe sideways to see the whole diagram.

Full bandwidth NDI. At about 150 Mbit/s for each 1080p60 camera (indicative), six cameras put about 900 Mbit/s on a 1 Gbit/s uplink. That leaves no room for anything else, and a burst drops frames.

NDI HX3. HX3 compresses with H.264 or HEVC to about 50 to 62 Mbit/s for 1080p60, so the same six cameras use about 370 Mbit/s. I check every device on the production supports HX3.

Only send what is needed. A camera only sends a stream to receivers that ask for it, and NDI can use multicast so several receivers share one stream. With IGMP (Internet Group Management Protocol) snooping, the switch sends each stream only where it is wanted.

Animation

Try it: is the upload big enough?

Pick the stream format and how many platforms the encoder sends to directly, then enter the upload speed measured at the venue. The tool uses YouTube's recommended bitrates and plans one and a half times the stream rate as headroom (typical).

1
Stream bitrate
17.1 Mbit/s per platform
Total sent
17.1 Mbit/s
Upload to plan for
about 26 Mbit/s
Line use
86% of the measured upload

Tight. One 1080p60 H.264 stream at 17 Mbit/s plus audio needs about 26 Mbit/s with headroom, and the line gives 20. I would drop to 1080p30 at about 14 Mbit/s, use a wired second line, or send one stream to a restreaming service that copies it to each platform.

Network rules I keep for streams

Wired, always

The encoder, the switcher and every NDI device go on cable. Wi-Fi is for the audience, never for the stream.

A production VLAN

Production traffic gets its own VLAN (virtual local area network), away from guest and office traffic, with its own addresses and no surprise firewalls.

Multicast done properly

If NDI uses multicast, the switches need IGMP snooping and a querier, or every stream floods every port. The AV over IP page walks through it.

Prioritise the stream

On a shared connection, QoS (quality of service) marks the outgoing stream so office traffic cannot starve it. Marking and queuing are on the QoS page.

Measure the real upload

I test the upload at the venue, at the time of the event, for several minutes. A speed test burst flatters a busy line.

A second way out

A second line, a bonded cellular unit or a phone tether ready to take over, plus a backup stream key the platform can fail over to.

Streams that cross a building network, with NDI, Dante and multicast, are on my AV over IP page.

Recording

Record everything, and plan for failure

A live stream happens once, but the recording lasts. I record the programme and every camera, and I plan what happens when each part of the chain fails, because one of them eventually will.

Record the programme

The switcher or OBS records the programme locally at a higher bitrate than the stream. If the stream fails, the event still exists.

Record every camera

ISO recording keeps each camera as its own file with matching timecode. ATEM ISO models also write a DaVinci Resolve project, so the edit can recut the show.

Record the sound separately

A multitrack recording from the mixer lets me remix the sound afterwards. A two track backup goes to a second device.

Two copies before leaving

Before I pack down, the media is copied to a second drive. The post production page covers what happens next.

Planning for things to fail

FailureWhat I prepare
Encoder or computer crashA second encoder, or the switcher's own streaming engine as a backup, with a backup stream key.
Internet dropA second line or bonded cellular, and a platform setting that keeps the event open for a few minutes.
Camera failureA wide safety shot that never moves, so there is always something to cut to.
Sound failureA backup microphone at the lectern and a second feed into the switcher.
Power cutA small UPS (uninterruptible power supply) on the switcher, encoder and network switch.

The ISO recordings become an editable timeline afterwards, as on my post production page.

Troubleshooting

Finding a fault along the chain

Most streaming faults sit in one of five places: the source, the cable, the switcher, the encoder and its network, or the platform. Choose a symptom to see where it usually sits and what I check first.

A live streaming fault finder Five places a streaming fault can sit, in order: the source, the cable, the switcher, the encoder and network, and the platform and players. For each symptom, test traffic flows as far as the place that fails. A panel shows what I see and what I check first. Sourcecamera, laptop CableHDMI, SDI Switchermix and sound Encoderand network Platformand players What I see What I check first Symptom: One camera shows black Input black; the camera's own screen is fineCamera output menu: match 1080p50 Symptom: Laptop video plays as black Slides show, but videos from apps go blackPlay the clip from a file, not an app Symptom: Picture flickers or rolls Dark bands roll up the picture under LED lightsShutter 1/50 or 1/100 at 25 fps Symptom: Sparkles and dropouts on SDI White dots, then the picture drops outSwap in a short known good cable Symptom: Sound out of step with lips Clap test: sound arrives before the pictureDelay the audio input by frames Symptom: Remote guest hears themselves The guest hears their voice back, lateCheck the mix minus bus on the mixer Symptom: Stream drops frames OBS stats: frames dropped by the networkSustained upload test, then lower bitrate Symptom: SRT link glitches SRT stats: lost and resent packets climbRaise SRT latency to several RTTs Symptom: Viewers buffer, health is good Platform says the stream is healthyCheck bitrate, keyframes, lower qualities

Swipe sideways to see the whole diagram.

One camera shows black on the switcher. Almost always a format mismatch: the camera sends 1080p25, 1080i or 4K while the switcher runs 1080p50. I set the camera's output to the switcher's format, or use an input with a standards converter.

Slides show but a video goes black. The laptop is protecting content with HDCP (High bandwidth Digital Content Protection), which capture and streaming kit is not allowed to show. I play the clip from a local file in a plain player, or put the video into the switcher directly.

The picture flickers or rolls. The camera shutter is fighting the 50 Hz mains lighting. I set the shutter to 1/50 or 1/100, or adjust it until the bands stop, and keep every camera the same.

Sparkles and dropouts on an SDI feed. The run is too long for the format or the cable, or a BNC connector or barrel is poor. I test with a short known good cable, check the reach for 3G or 12G, and add a reclocking amplifier or fibre for long runs.

Sound is out of step with the picture. Video is delayed by cameras and converters while sound arrives at once. I measure with a clap or a sync clip and delay the sound input on the switcher by the matching number of frames, then check on the stream itself.

A remote guest hears their own voice. They are being sent the full programme sound. I send them a mix minus: everything except their own voice.

The stream drops frames. Dropped frames in OBS mean the network cannot keep up; skipped frames mean the computer cannot encode in time. I test the sustained upload, lower the bitrate or the resolution, use a hardware encoder and keep the encoder on a cable.

An SRT link glitches every few seconds. Resent packets arrive after the latency buffer has played out. I look at the round trip time and loss in the SRT statistics and raise the latency to several round trips, or find a cleaner connection.

Viewers buffer while the stream health is good. Viewers' connections cannot take the full bitrate, the keyframe interval is too long, or the platform is not offering lower qualities. I keep keyframes at 2 seconds, use a sensible bitrate and check the platform makes lower quality copies.

Animation

Where I look

QuestionWhere I lookWhat it tells me
Is every source in the right format?The switcher's software, such as ATEM Software Control, and the multiviewWhich inputs are locked, and their format
Is the network or the computer the problem?OBS, View, StatsFrames dropped by the network versus frames skipped by the encoder
Is the stream arriving cleanly?The platform's stream health page, such as YouTube Live Control RoomBitrate received, keyframe interval and warnings
Is the internet link clean?SRT statistics on the encoder and decoderRound trip time, lost and resent packets, and buffer use
What does the line really give?iperf3 -c server -R -t 60 or a sustained upload testUpload over a full minute, not a short burst
Is sound in sync?A clap on camera, or a sync test clip played through the chainHow many frames to delay the sound

I change one thing at a time and test again on the stream itself, from a phone on mobile data, not just on the production monitors. My general method is on the troubleshooting page.

Engagement

How I plan a live stream, from first call to the last upload

A stream looks simple to viewers and is anything but for the crew. I follow the same path for every stream, small or large. I agree who is watching and why, set clear targets and visit the venue. Then I plan the cameras and the network on paper, rehearse the whole chain, run the show calmly and finish with the recordings.

How an engagement runs, from first conversation to lifecycle Seven stages from left to right: engage (the brief), requirements (FR and NFR), discovery (site visit), design (plan), build and test (rehearse), support (show day), lifecycle (after). For each stage the panel shows who I work with and what comes out of it. After the lifecycle stage the loop returns to the start. Engagethe brief RequirementsFR and NFR Discoverysite visit Designplan Buildrehearse Supportshow day Lifecycleafter Step 1 · EngageAgree who watches, where and on whatWho I work withWhat comes outEvent organiserPresenters and speakersVenue managerAudience and platformsShow running orderRights and consentsBudget and crew Step 2 · RequirementsTargets for picture, sound, delay and backupWho I work withWhat comes outEvent organiserPlatform ownerVenue ITFormat and bitrateDelay target for the streamBackup and recording needsCaptions and accessibility Step 3 · DiscoveryUpload, power, distances and sight linesWho I work withWhat comes outVenue ITVenue technicianEvent organiserMeasured upload at show timeCable runs and distancesPower and camera positionsNetwork access and VLAN Step 4 · DesignCamera plan, signal flow and networkWho I work withWhat comes outDirectorSound engineerVenue ITCamera plan and shot listSignal flow and cable scheduleStream settings and keysFailure plan Step 5 · Build and testTest the whole chain before doors openWho I work withWhat comes outCrewPresentersRemote sitesLine and camera checksPrivate test streamLip sync checkedRemote links proven Step 6 · SupportWatch the stream, the sound and the chatWho I work withWhat comes outDirectorStream producerVenue ITStream health watchedComms and cue callsIncident logBackup ready to switch Step 7 · LifecycleRecordings, edits, lessons and upgradesWho I work withWhat comes outEvent organiserEditorKit ownerMedia copied twiceEdited version deliveredLessons learnedFirmware and kit updates

Swipe sideways to see the whole diagram.

Step 1
  1. I start with who will watch and how: a public stream, a private link or only a recording for later. I agree the platforms, the running order, music and image rights, consent to film the audience, and what the organiser can spend on cameras, crew and connectivity.

  2. I turn the brief into requirements: the format and bitrate, how much delay is acceptable, what must be recorded, and what happens if the internet fails. Accessibility counts too, with captions for viewers who need them. Each target has an owner who agrees it.

  3. I visit the venue at the same time of day as the event. I measure the upload over several minutes, walk every cable run, find power for each camera position, check sight lines, and agree a wired network port or VLAN with the venue's IT team.

  4. On paper I draw the camera plan and shot list, the signal flow from every source to the stream, and a cable schedule with lengths and types. I set the stream format and keys, and write down what happens when each part fails.

  5. I build early and test everything: each camera and microphone, graphics, a private test stream watched on a phone on mobile data, lip sync, and every remote link. Presenters rehearse their entrances, slides and handovers so nothing surprises them live.

  6. During the show the director calls shots, and someone watches the stream itself, its health page and the viewer chat. I keep an incident log, so problems are fixed calmly and remembered, and the backup line and encoder stay ready to take over.

  7. After the show the media is copied twice before anything is deleted, the edited version goes to the organiser, and we review what went well and what did not. Between events I update firmware and software, test it, and replace worn cables.

Functional requirements

What it must do.

  • Stream the programme to the agreed platforms, publicly or privately, from the planned start until the end.
  • Switch between every planned camera, the slides and the graphics, with names on screen for each speaker.
  • Carry a clean stereo sound mix, in sync with the picture, with every speaker clearly audible.
  • Bring in remote presenters or sites, with a return feed and talkback for each.
  • Record the programme, and every camera where an edit is planned, at a higher quality than the stream.
  • Keep a holding slide ready for the start, the breaks and any failure.

Non functional requirements

How well it must do it, with a number wherever one can be agreed.

Picture
1080p at 25 or 50 fps, at the platform's recommended bitrate (typical).
Delay
Viewers see the stream within the platform's normal or low latency setting, agreed in advance.
Lip sync
Sound within the EBU R37 window: no more than 40 ms early or 60 ms late.
Upload
A measured, sustained upload of at least one and a half times the total stream bitrate (typical).
Resilience
No single failure stops the stream for more than a minute (typical), with a backup line and encoder.
Recording
Two copies of every recording before the crew leaves the venue.
Accessibility
Captions available on the stream or added to the recording afterwards.
Security
Stream keys kept private, remote links encrypted, and production devices off the guest network.

Numbers are typical starting points. Each event agrees its own with the organiser and the venue.

Projects

Where I use it

I stream and record video in my own time. My network work gives them a solid base, because a modern production is a network as much as a set of cameras. These cards show the kit I use, the training I have done and the projects behind it.

Hobby · ongoing

Multi camera streams with ATEM switchers

In my own time I stream and record with Blackmagic ATEM switchers: the ATEM Mini series for HDMI sources, and a larger ATEM with SDI inputs for longer cable runs.

Hobby · ongoing

PTZ and cinema cameras

My cameras are PTZ cameras for fixed angles under remote control, and cinema cameras for the main shots. Making them match when I cut between them is most of the work.

Hobby · ongoing

Sound from my own mixer

I mix sound on an Allen & Heath Qu series mixer. The mixer side, from microphones to the stream feed, is on the live sound page.

Kramer Academy · courses completed

AV fundamentals and AV over IP

I have completed Kramer's AV Fundamentals, AV over IP Installer and AV over IP Sales Associate courses, which cover carrying audio and video across networks.

Home lab · ongoing

Real time media on a network

My lab carries voice calls between three sites over an MPLS (Multiprotocol Label Switching) core, with a voice VLAN on every phone port. Live video needs the same care: bandwidth, separation and priority. See the topology.

JJ Food Wholesale · 2007 to 2011

Voice over IP across a group

LAN (local area network), WAN (wide area network), VoIP (voice over IP), virtualisation and security across the group, including the warehouse and cold chain systems at the food depots.

Backed by Kramer's AV Fundamentals and AV over IP courses, and CCVP (Cisco Certified Voice Professional) for real time media on networks. See all credentials